Domain Pipelines

These packages give AI agents a long-term memory layer that persists information across sessions, not just within a single conversation. You'd adopt them to solve the problem of agents forgetting user preferences, contradicting themselves, or accumulating irrelevant information over time. The most distinctive thing is that memory is treated as a managed asset with an explicit lifecycle—extraction, decay, forgetting, and contradiction resolution—rather than a fire-and-forget vector store.

These JSON files give you structured data on bicycle brands, models, and sizes extracted from over 25,000 real-world listings. You would use them to populate a bike eCommerce catalog, build a bike recommendation system, or standardize bike model data across applications. Each brand is stored in its own file, with every model listing available sizes, frame measurements, and suggested rider height ranges.

These packages give you a complete, modular RAG stack built around Qdrant, with hybrid retrieval that combines vector search, BM25 keyword search, and cross-encoder reranking. You'd adopt them to avoid assembling and integrating a dozen separate tools for document ingestion, chunking, retrieval, reranking, evaluation, and observability into a coherent pipeline. The most distinctive thing is that every component—from four chunking strategies to ablation studies to an MCP server with 41 tools—is a separate, independently installable package sharing core types and Zod schemas, so you can use only what you need while keeping everything type-safe and composable.

This is a desktop invoicing application built with Electron, React, and TypeScript that manages customers, products, and invoices with email delivery and PDF generation. You would use it to handle invoicing for a freelance or contracting business without paying subscription fees or per-invoice charges. The application stores all data locally in SQLite, generates PDFs via Puppeteer with Mustache templates, and sends invoices through SMTP email—all running as a standalone desktop app with no external service dependencies.

These packages give you a set of MCP tools for generating and processing images, audio, video, documents, and 3D models, then chaining those operations into pipelines with artifact passing between steps. You would adopt them to avoid wiring together multiple provider SDKs, retry logic, cost tracking, and caching yourself when building AI agents that produce media as output. The most distinctive thing is that every operation is a composable pipeline step with built-in quality gates, budget enforcement, content-addressed caching, and multi-provider routing, so you can define a multi-step workflow once and have it handle provider fallback, cost caps, and resumable execution automatically.

These packages give you the full pipeline to turn microphone or phone audio into an AI agent response and back — speech-to-text, MCP tool calls, text-to-speech, and telephony or WebRTC transport, all orchestrated with per-stage latency budgets. You'd adopt them to build a production voice agent without writing the audio plumbing, provider switching, or session management yourself. The most distinctive thing is that the agent logic lives entirely in an external MCP server, so the pipeline is a pure transport layer that can be swapped between Twilio, WebRTC, or a local simulator without changing the agent.